AI assistants
Put your assistant where your users already talk.
message.received comes in, a typing indicator goes out while your model thinks, and the answer goes back in the same thread, on WhatsApp and Instagram.
Great assistants die in a chat widget nobody opens.
Users don’t come back to your site
They’re in WhatsApp all day; your widget waits for a visit that doesn’t happen.
Silence feels broken
A model that takes ten seconds without a signal reads as a failed send.
Every channel is a new integration
WhatsApp and Instagram each have their own APIs, payloads and webhooks.
Three calls for a conversational loop.
Receive, show typing, reply. The same three calls on every live channel.
The whole loop
Inbound message to answer in a handful of lines.
In the user’s own thread
Your assistant answers from your customer’s business number, in the chat the user started.
Assistants that do the work.
Connect the model to your tools, and the conversation to WhatsApp.
Answer product questions
Ground answers in your catalogue and reply with text or an image.
Take actions in your system
Let the assistant look up orders or book appointments, then confirm in the chat.
Hand off to a human
Route a conversation to your team and reply from the same API.
Read receipts
Mark messages read so users know they were seen.
Signed inbound
Verify every inbound event before your model acts on it.
Production-ready from the first reply.
What a conversational product needs underneath.
Retries with backoff
Inbound events are retried if your endpoint is busy, and every attempt is logged.
Sandbox first
Test the whole loop with test keys and registered test recipients.
Rate limits you can see
429s carry Retry-After so your worker knows exactly when to try again.
Frequently asked questions
Can’t find yours? Ask on Discord or email support@intelliconcierge.com.
Which model can I use?
Any. Intelli delivers the messages; your code calls whichever model you choose.
Can the assistant message people first?
On WhatsApp, outside the 24-hour service window you need an approved template. Inside it, reply freely.
How do I show the assistant is thinking?
POST /messages/typing with the client_ref and recipient while you generate the answer.
Ship it this week.
Get a key, test in the sandbox, and connect a real number when you’re ready.